Active Learning Sampling for Material Structure and Energy Diversity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional active learning methods struggle to accurately predict energy values in materials due to difficulties in sampling training data that are similar in structure but differ in energy, leading to reduced accuracy in energy prediction using neural networks.
Innovation Solution
An active learning method that uses a neural network to sample training data based on both structural features and energy prediction values, considering the diversity of both structure and energy, thereby enhancing the accuracy of energy prediction while reducing the number of required training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional active learning methods sample training data based only on structural features, then the sampling process is simple, but the energy prediction accuracy is reduced because data with similar structure but different energy cannot be properly distinguished
Solution Approach 1:
The patent merges structural features and energy prediction values into a unified feature space for sampling. By combining these two types of features, the method can simultaneously consider both structural similarity and energy differences when selecting training data, thereby improving energy prediction accuracy while maintaining a relatively simple sampling framework.
Solution Approach 2:
The patent adds the energy prediction value as an additional dimension to the traditional structural feature space. This dimensional extension allows the sampling method to distinguish between materials with similar structures but different energies, enabling more accurate energy prediction while keeping the sampling approach conceptually straightforward.
2Measurement precision
If a large amount of labeled training data is prepared to improve energy prediction accuracy, then the prediction performance improves, but the computational workload and cost of DFT simulation increase significantly
Solution Approach 1:
The patent employs active learning where the neural network itself identifies and selects the most informative training data samples. The model serves its own training needs by autonomously determining which unlabeled data points would be most beneficial to label, thereby reducing the overall amount of labeled data required while maintaining high prediction accuracy and lowering computational costs.
Solution Approach 2:
The patent changes the sampling strategy from random or structure-only based selection to a hybrid approach that incorporates both structural features and energy prediction values. This parameter change in the sampling criteria enables more efficient selection of training data, reducing the total quantity needed while improving the quality and representativeness of the training set.
3Adaptability or versatility
If training data with diverse structures is used to capture material diversity, then the model generalization improves, but the energy diversity within similar structures may be insufficient leading to poor energy prediction
Solution Approach 1:
The patent combines structural features and energy prediction values into a unified sampling criterion. This merging allows the selection process to simultaneously ensure structural diversity (for generalization) and energy diversity (for accurate energy prediction), resolving the conflict between these two requirements by considering both aspects in a single integrated framework.
Data Source
AI summary
A computer-readable recording medium has stored therein an active learning program for causing a computer to execute a process including: extracting a first feature related to a structure of each of a plurality of materials by inputting a plurality of structure data associated one with each of the plurality of materials into an active learning neural network; obtaining a second feature related to energy of the structure of each of the plurality of the materials, using the active learning neural network, the second feature being based on the first feature; and determining, based on the first feature and the second feature of each of the plurality of materials, one or more structure data to be training data for training an energy prediction neural network for predicting energy of a material from among the plurality of structure data.